[Springframework-developer] CGLIB question

[Rod J. <rod...@in...>]

Chris,

I'm sure you can help with this one.

I guessed that the CGLIB MethodFilter interface was what I needed to get
CGLIB override to get only those methods with advice overloaded. However,
when I do that I find that behavior changes.

Here's how I've changed the inner classes in
org.springframework.aop.framework.AopProxy. I want only methods that have an
advice chain to be proxied.


 private class CglibProxyFactory {

  private Object createProxy() {
   try {
    return Enhancer.enhance(advised.getTarget().getClass(),
     completeProxiedInterfaces(),
     new MethodInterceptor() {
      public Object intercept(Object handler, Method method, Object[]
objects, MethodProxy methodProxy) throws Throwable {
       return invoke(handler, method, objects);
      }
     },
// I've added the three following arguments
     null,  // ClassLoader: use default
     null, // Method replace: what does this mean? I couldn't work out from
CGLIB Javadoc
     new SelectiveOverrideMethodFilter()
    );
   }
   catch (CodeGenerationException ex) {
    throw new AspectException("Couldn't generate CGLIB subclass of class '"
+ advised.getTarget().getClass() + "': " +
      "Common causes of this problem include using a final class, or a
non-visible class", ex);
   }
  }
 }


 private class SelectiveOverrideMethodFilter implements MethodFilter {

  public boolean accept(Member member) {
   // Proxy is not yet available, but that shouldn't matter
   //List chain =
advised.getAdvisorChainFactory().getInterceptorsAndDynamicInterceptionAdvice
(advised, null, (Method) member, advised.getTarget().getClass());
   //return !chainIsOptimizableToDirectTargetInvocation(chain);
   return true;
  }

 }

The two lines I've commented out in the accept() method produce a 2.5x
performance improvement in non-advised methods (as you'd expect) but they
also produce different behavior, compared to returning true.

It appears that now CGLIB is creating a new instance of the target and
somehow now copying its state. Previously a property I'd set on the target
was visible through the proxy; now it isn't.

Overall I really like CGLIB: it does what I expect of it with no fuss, and
it's a cool concept. But the Javadoc really is inadequate.

Regards,
Rod